This paper presents a method using kernel Canonical Correlation Analysis (CCA) to learn a semantic representation for web images and their associated text. The semantic space enables comparison between text and images, facilitating content-based retrieval without explicit labeling. The authors compare two approaches for retrieving images based on their content from text queries, using the Generalized Vector Space Model (GVSM) as a benchmark. They explore the use of partial Gram-Schmidt orthogonalization and incomplete Cholesky decomposition to reduce computational complexity and improve performance. The proposed method is validated through experiments on a multimedia image-text database, demonstrating superior performance in both content-based and mate-based retrieval tasks. The paper also discusses the selection of the regularization parameter and provides a generalization framework for CCA, extending it to more than two sets of variables while preserving its key properties.This paper presents a method using kernel Canonical Correlation Analysis (CCA) to learn a semantic representation for web images and their associated text. The semantic space enables comparison between text and images, facilitating content-based retrieval without explicit labeling. The authors compare two approaches for retrieving images based on their content from text queries, using the Generalized Vector Space Model (GVSM) as a benchmark. They explore the use of partial Gram-Schmidt orthogonalization and incomplete Cholesky decomposition to reduce computational complexity and improve performance. The proposed method is validated through experiments on a multimedia image-text database, demonstrating superior performance in both content-based and mate-based retrieval tasks. The paper also discusses the selection of the regularization parameter and provides a generalization framework for CCA, extending it to more than two sets of variables while preserving its key properties.